Our first lecture meeting has just been held, and I’m glad to cautiously say that it was a success.
We were able to film a general introduction to the shutdown problem and cover most of the corrigibility paper of Soares et al. (2015).
What we’ve learned:
– even if the first attempt is already okay, we will want to have (at least) 2 attempts for each shot, because the second attempt is invariably going to be substantially better.
– it is essential for the lecturer to have some way to engage with the content while lecturing. We used a whiteboard this time, but we thought of the setup with a wacom and a projector that Remmelt sketched above. It’s better in intuitive ways.
– we need to add video editing as a full role. It’s more labor-intensive than expected.
– switching backgrounds between high-level explanations and math explanations might a good way to help students switch between subagents to better parse very divergent content
What is to happen next:
– Chris is going to do a first pass of video editing, cropping parts and cutting out the failed bit.
– someone needs to take care of hosting and set up an instance of EdX.
– then we put up the videos, and we will want various people to watch them and review them as critically as possible
– we need someone to watch the videos and insert assignments between them
– an appointment is going to be made for the next lecture meeting, somewhere in the last 2 weeks of July. We wanted to stick to just corrigibility until we were sure we got the setup exactly right before filming other subjects. However, Johannes Treutlein from FRI will come to the Netherlands to give a guest lecture, so we will cover decision theory just this once instead.